Why most startups are building AI the wrong way
AI isn’t software. And that’s why most founders are building it wrong.
If you’d like to sponsor the Product Market Fit, email me at g@guillermoflor.com
Three kids sleeping on the floor in San Francisco are replacing accountants using AI.
Meanwhile, most startups still don’t know how to build with AI.
And the reason is simple. They are treating AI like it’s just better software.
In this piece, I want to break down three ideas:
AI is not software
Industry insiders are often the worst people to build with it
Outsiders keep winning for a reason
1. AI Is Not Software
Most founders are taking old software workflows and plugging in an LLM.
They automate steps. They rewrite UI. They add chat.
And then they wonder why the product still feels broken.
AI allows you to completely rethink the path between problem and solution.
You no longer need step-by-step flows.
You no longer need the user to click through menus and logic trees.
You just need to ask what the desired outcome is and let the system figure out the rest.
If your product still looks like the old one, you’re probably doing it wrong.
2. Industry Experts Bring Baggage
People with deep experience in a space tend to recreate what already exists.
They respect legacy constraints. They follow invisible rules.
And they end up rebuilding old tools with a layer of AI on top.
This is why so many AI tools feel like slower, more confusing versions of the thing they are meant to replace.
Outsiders don’t carry that weight.
They are not trying to preserve a system.
They are trying to replace it.
Peter Zhu and Om Agarwal didn’t work in finance.
They just saw how broken startup accounting was and decided to rebuild it using AI.
Their company, Minerva, is doing everything from transaction classification to reporting without copying the logic of existing software.
That’s why it works.
3. Outsiders Keep Winning
Shopify started because Tobi wanted to sell snowboards and the tools available were unusable.
Canva began when Melanie Perkins saw that most people just needed a simple way to design, not another pro tool.
Spotify worked because Daniel Ek understood that piracy wasn’t about money, it was about speed and access.
The same thing is happening now in AI.
And once again, the biggest breakthroughs are coming from people who are not held back by experience.
If You’re Building With AI
Don’t start by copying an existing product.
Start by asking what should exist if you had no constraints.
Then use AI to make that possible.
The biggest winners are not adding AI to software.
They are replacing software entirely.
PS:
ez Ibar and I are looking for a videographer who can help us tell these stories properly. If this sounds exciting, send us a message!Cheers,
-Guillermo